I’m currently working on an image recognition project, and the pre-processing steps before the model execution take a considerable amount of time. When an error occurs in the model, I have to re-run everything from the beginning, which is very time-consuming. Is there a way to avoid running the entire process from scratch and only execute the model part?